3D mapping: light ground reflection and multiple walls

classic Classic list List threaded Threaded
4 messages Options
acp
Reply | Threaded
Open this post in threaded view
|

3D mapping: light ground reflection and multiple walls

acp
This post was updated on .
Dear people.


I do have a ZED camera and a Jetson TX2 mounted on a drone. I have taken the drone with my hands to map an outdoor place by means of the rtabmap_ros node.


I have been trying to map big areas where I got some distortions during mapping as you can see in this link:

http://official-rtab-map-forum.206.s1.nabble.com/rtabmap-ros-outdoor-3D-mapping-td6104.html#a6117

However, I made a very simple test. I have put the drone on a wheel chair and i turned the chair to complete a circle.


And I can see the following.

* when the zed is looking at the window it detects the the reflected light on the floor and plots the floor  under the ground in RVVIZ.


* When the zed  is turning it detects multiple walls.







I share a rtabmap.db just in case :)

http://www.fit.vutbr.cz/~plascencia/rtabmap_lab.db 


I just calibrated the zed camera by means of matlab and tested the calibration by means of opencv, The calibration looks good :)


If you need I can share the launch file or if there is any parameter i need to tune or what can i do to have a good 3D mapping.


I am looking forward to your replay.



 
Reply | Threaded
Open this post in threaded view
|

Re: 3D mapping: light ground reflection and multiple walls

matlabbe
Administrator
Hi,

looking at your screenshots, the depth distortion is coming from the disparity algorithm used from ZED sdk. Featureless areas will appear distorted because the algorithm interpolates them. You could compare with rtabmap using stereo images directly from zed camera (instead of using the depth image created by zed sdk), and see how the depth image is created (using cv::StereoBM). There could be also some parameters on zed camera driver side that could be tuned to avoid the algorithm to interpolate too much the featureless surfaces.

cheers,
Mathieu
acp
Reply | Threaded
Open this post in threaded view
|

Re: 3D mapping: light ground reflection and multiple walls

acp
Hi Mathew,

Thank you very much for your answer,

and, how about the multiple walls detected? how can be solved that issue?


Regards
Reply | Threaded
Open this post in threaded view
|

Re: 3D mapping: light ground reflection and multiple walls

matlabbe
Administrator
As I refered in my previous post, you could try strereo input of zed instead of RGB-D to recompute de depth images with a different algorithm, or tune the zed parameters (if possible) to avoid estimating depth on featureless areas.